面向SAR图像目标分类的CNN模型可视化方法

卷积神经网络(CNN)在合成孔径雷达(SAR)图像目标分类任务中应用广泛。由于网络工作机理不透明,CNN模型难以满足高可靠性实际应用的要求。类激活映射方法常用于可视化CNN模型的决策区域,但现有方法主要基于通道级或空间级类激活权重,且在SAR图像数据集上的应用仍处于起步阶段。基于此,该文从神经元特征提取能力和网络决策依据两个层面出发,提出了一种面向SAR图像的CNN模型可视化方法。首先,基于神经元的激活值,对神经元在其感受野范围内的目标结构学习能力进行可视化,然后提出一种通道-空间混合的类激活映射方法,通过对SAR图像中的重要区域进行定位,为模型的决策过程提供依据。实验结果表明,该方法给出了模型在不同设置下的可解释性分析,有效拓展了卷积神经网络在SAR图像上的可视化应用。     

基于共轭梯度法的CNN图像分类模型研究

本文针对处理图像分类的卷积神经网络(CNN)模型,设计并应用修正的三项PRP共轭梯度法(M-PRPCG)训练模型以提高图像分类的准确率。首先,基于CIFAR-10图像分类数据集构建ResNet18和VGG16卷积神经网络(CNN)模型;然后在训练集上,采用修正的三项PRP共轭梯度法(M-PRPCG)训练模型;最后在测试集上进行验证。实验结果表明,修正的三项PRP共轭梯度法(M-PRPCG)训练的ResNet18和VGG16模型相比于Adam算法训练的模型在图像分类准确率上分别提高了3.46%和1.97%。     

车牌图像分类识别技术研究

在现阶段的交通管理领域,普遍应用车牌识别系统是交通信息化的一个重要组成部分。为了提高车牌图像识别技术应用的效果,文章围绕车牌图像分类识别技术做出分析,在保证车牌图像识别精度基础上提高识别的效率。本文首先介绍车牌图像分类识别技术,了解该技术基本情况;其次介绍车牌图像采集技术、车牌图像特征值提取与分类器、车牌图像处理技术3种车牌图像分类识别的常见技术,了解不同技术在车牌图像分类识别中的应用要点;最后提出加大采集图像内容与质量控制力度、建立车牌识别样本数据库、明确车牌图像识别规范3点建议,明确今后车牌图像分类识别技术的发展方向,以期能够为今后车牌图像分类识别的发展夯实基础。     

深度学习结构优化的图像情感分类

自然图像情感分类在分析用户需求、监控网络舆情等方面具有重要意义。然而基于深度学习的分类算法存在训练过程难以控制、分类结果缺乏解释的问题。为此提出一种人类知识驱动的深度学习结构优化算法。首先通过特征可视化显示卷积神经网络提取的情感特征;其次结合人类对图像情感可视化结果的感知来优化网络结构,利用人类知识驱动网络,重点学习情感信息更明显的特征;最后对所构建网络的参数进行微调,使其更适用于自然图像情感分类任务。在Twitter情感图像数据集上与其他分类方法的对比实验表明,所提出的算法获得了88.1%的分类准确率,优于其他方法。消融实验证明网络优化结果比未优化提高了8.1%。类激活图、空间位置和神经元组特征可视化直观解释了模型运作的过程与原因,进一步证实算法识别自然图像情感的能力。     

基于CNN的乳腺癌病理图像分类研究

乳腺癌已经成为全球第一大癌症,乳腺癌的早期发现及良恶性诊断对于治疗具有重要的意义.针对传统机器学习方法在乳腺癌病理图像分类任务中性能不足和准确率低的问题,本文提出了基于CNN(卷积神经网络)的乳腺癌病理图像分类模型,将乳腺癌病理图像分为良性与恶性.该模型以VGG网络为基础,对网络结构进行调整,在公开的BreakHis数...     

深度学习算法的激活函数研究

作为深度学习算法中重要的环节,激活函数可以为神经网络引入非线性因素.大量学者通过提出或改进激活函数的方法在一定程度上提高了算法的优化及泛化能力.研究了现阶段的激活函数,将激活函数大致分为S系激活函数和ReLU系激活函数,从不同激活函数的功能特点和存在的饱和性、零点对称和梯度消失及梯度爆炸的现象进行研究分析,针对Sigm...     

基于CNN和RFC的极化SAR图像分类

卷积神经网络(Convolution Neural Network,CNN)在极化合成孔径雷达(Synthetic Aperture Rader,SAR)图像分类的应用中在类别非边界区域取得了好的分类结果,在类别边界区域没有取得好的分类结果,随机森林分类器(Random Forest Classifier,RFC)在极...     

基于分块CNN的多尺度SAR图像目标分类算法

针对合成孔径雷达（Synthetic Aperture Radar, SAR）图像目标分辨率差异大,多尺度SAR图像目标分类准确率不高的问题,提出了一种基于迁移学习和分块卷积神经网络（Convolutional Neural Network, CNN）的SAR图像目标分类算法。首先通过大量与目标域相近的源域数据对分块CNN的参数进行训练,得到不同尺度下的CNN特征提取网络;其次将CNN的卷积和池化层迁移到新的网络结构中,实现目标特征的提取;最后用超限学习机（Extreme Learning Machine, ELM）网络对提取的特征进行分类。实验数据采用美国MSTAR数据库以及多尺度SAR图像舰船目标数据集,实验结果表明,该方法对多尺度SAR图像的分类效果优于传统CNN。     

融合CNN与Transformer结构的遥感图像分类方法

为解决高分辨率遥感图像所具有的类内差异大而类间差异小的特性导致的图像难分类问题,提出一种基于深度学习中卷积神经网络与Transformer优点的混合结构。对卷积层提取的特征信息使用两个带有空间位置信息的注意力机制,分别沿水平方向和垂直方向对每个通道进行特征聚集,以减少遥感场景特征的冗余映射,使网络能够提取更多与任务目标相关的信息。然后利用Transformer编码器结构对捕获的特征图进行编码操作,赋予特征图中感兴趣区域较大的权重。实验结果表明,与现有的基于深度学习的遥感图像分类方法相比,所提方法既降低了模型参数量,又提升了分类准确率,在遥感图像分类数据集AID、NWPU-RESISC45及VGoogle上均达到了最高的平均分类准确率,分别为98.95%、96.00%和95.01%。     

卷积神经网络中ReLU激活函数优化设计

卷积神经网络中的激活函数的作用是激活神经元的特征然后保留并映射出来,这是神经网络能模拟人脑机制,解决非线性问题的关键。ReLU函数更是其中的佼佼者,但同时其自身也存在不足之处。文章从两个方面对ReLU函数进行了优化设计。对使用梯度下降法的激活函数的学习率进行讨论研究并提出可行的学习率改进方法。提出一种新型校正激活函数,称其为e-ln函数,经过Mnist数据集仿真实验证明某些情况下其性能要优于ReLU。     

TheiaNet: Towards fast and inexpensive CNN design choices for image dehazing

This work examines inexpensive design choices for dehazing as an end-to-end image-to-image mapping problem without relying on the physical scattering model. The proposed TheiaNet is free from intermediate-computation of transmission map, enabling haze removal in a highly resource constrained environments. The simplicity of the network is augmented by a spatial cleaning bottleneck block, that adds faster feature extraction without adding to trainable parameters. We also analyze the effectiveness of multi-cue color space (RGB, HSV, LAB, YCbCr) over single cue color space (RGB) for end-to-end dehazing. A comprehensive set of experiments were conducted on HazeRD, D-Hazy and the more recent Reside datasets. The proposed TheiaNet significantly outperforms the existing CNN and GAN based state-of-the-art methods in terms of PSNR and SSIM on all these datasets. It also outperforms all existing methods in term of speed, compute and memory efficiency, making it more efficient. This work highlights how judicious application-specific components can augment simple CNNs to denoise faster, and more accurately than deeper heavier networks, which is supported by an ablation analysis as well.     

An efficient feature fusion in HSI image classification

Multidimensional Systems and Signal Processing - In recent times, the fusion of spatial relaxation with spectral data has achieved remarkable success in target classification methods. Spatial...     

高效率LED驱动电源设计

随着LED生产成本下降,越来越多应用开始采用这类组件,包括手持装置、汽车电子和建筑照明等。LED拥有高可靠性、良好效率和超快响应速度,所以很适合作为照明光源。虽然白炽灯泡的成本很低,更换费用却可能很昂贵。街灯就是很好的例子,更换一个故障灯泡往往需要出动多位人员和一辆卡车。也因为如此,尽管LED和白炽灯泡的效率大致相等,许多街灯却采用可靠性更高且更省电的LED。     

基于CNN池化和进化策略的一般神经网络图像分类研究

网络的爆炸式发展产生了海量的图像,图像标签的错误和缺失比较常见,图像分类研究很有必要。CNN池化能够提取到输入矩阵的重要特征,降低数据的维度。进化策略是模仿生物"优胜劣汰"进化方式的一种启发式算法,能快速找到问题的解。本文基于CNN池化提取一组有正确标签的图像的特征,搭建层数为3的神经网络,进化策略优化初始权重,通过训练集训练分类模型,通过测试集来验证模型的优劣,并使最终的模型实现对未知类别图像的高效分类。实例验证阶段收集10类100张犬类图片,按照各研发步骤进行实验,算法结果验证了进化策略优化权重的必要及神经网络模型的高效。     

基于视觉显著和CNN的花卉图像检索软件

花卉图像检索是图像检索领域的热门研究方向,高效、快速地检索数据库中的花卉图像一直是该方向的重点课题。为了检索花卉图像,文中设计了一个基于视觉显著模型和CNN的图像哈希算法,并根据此算法设计和开发出一个高效、快速的花卉图像检索软件。软件具有查询花卉类别、检索相似花卉、浏览花卉信息等功能。     

基于CNN+LSTM的改进心音分类模型研究

心脏听诊是心脏相关疾病提前诊断和筛查的重要手段,利用深度学习模型进行心音分类取得了不错的效果,但其分类度仍有待提高.该文在心音处理流程和模型结构两个方面做了优化处理,流程方面在心音分割后添加了归一化处理这一步,使得不同音频周期的心音放在了同一范围下比较,特征提取上使用了二阶谱分析方法,保留了提取特征的更多信息;在模型结...     

Multifeature extracting CNN with concatenation for image denoising

Convolutional neural networks (CNNs) have made great achievements in the field of image denoising but can still be improved. We introduce a network structure, namely, multifeature extracting CNN with concatenation (McCNN), which can preserve the edge and detail information and make the denoised image easier to view. The McCNN uses different-sized convolutional kernels to extract multiple features from the input image and send them into a forward network structure after cascading these features. The forward network structure consists of five nonlinear mapping modules, which are responsible for extracting more detailed textures and other advanced features. A skip connection is integrated into the forward network structure to pass the feature maps that carry many image details, which helps to reduce image distortion. The skip connection can also reduce gradient disappearance and improve network convergence speed. The potential clean image in the contaminated image contains much more information than the noise image. The noise image is regarded as the learning objective of the network to reduce the learning burden. The experimental results demonstrate that our McCNN denoising method can effectively remove Gaussian noise in grayscale images and offers objective and subjective quality improvement compared to that of the DnCNN-S, SCNN, and DSNet models, as well as other state-of-the-art denoising methods.     

CCNet: CNN model with channel attention and convolutional pooling mechanism for spatial image steganalysis

Image steganalysis based on convolutional neural networks(CNN) has attracted great attention. However, existing networks lack attention to regional features with complex texture, which makes the ability of discrimination learning miss in network. In this paper, we described a new CNN designed to focus on useful features and improve detection accuracy for spatial-domain steganalysis. The proposed model consists of three modules: noise extraction module, noise analysis module and classification module. A channel attention mechanism is used in the noise extraction module and analysis module, which is realized by embedding the SE(Squeeze-and-Excitation) module into the residual block. Then, we use convolutional pooling instead of average pooling to aggregate features. The experimental results show that detection accuracy of the proposed model is significantly better than those of the existing models such as SRNet, Zhu-Net and GBRAS-Net. Compared with these models, our model has better generalization ability, which is critical for practical application.     

Kernel eigenmaps based multiscale sparse model for hyperspectral image classification

Hyperspectral imaging (HSI) is the emerging method that combines traditional imaging and spectroscopy to provide the image with both the spatial and spectral information of the object present in the image. The major challenges of the existing techniques for HSI classification are the high dimensionality of data and its complexity in classification. This paper devises a new technique to classify the HSI named Spatial–Spectral Schroedinger Eigen Maps based Multi-scale adaptive sparse representation (S²SEMASR). In this, two different phases are employed for the accurate classification of the HSI, namely, Schroedinger Eigen maps (SE) based spatial–spectral feature extraction and multi-scale adaptive sparse classification for the feature extracted image. SE makes use of spatial–spectral cluster potentials which allows the extraction of features that best describes the characteristics of different classes of HSI. The multiscale adaptive sparse representation (MASR) applied over the SE features provides the sparse coefficients that includes distinct scale level sparsity with same class level sparsity. With the obtained coefficients, the class label of each pixel is determined. The proposed HSI classifier well utilizes the spectral and spatial characteristics to exploit the within-class variability and thus reduces the misclassification of similar test pixels Experimental results demonstrated that the proposed S²SEMASR approach outperforms the traditional results both qualitatively and quantitatively with an overall accuracy of 98.3%.